菜单
Mandatory external architectural and code-review harness for non-trivial code changes, PR reviews, refactors, modeling logic, dashboard/API integration, performance-sensitive work, and implementation planning. Must be followed by the code-hardener skill before Gemini edits production code.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
META-SKILL — mandatory entry point for EVERY non-trivial prompt. Decomposes prompt into 5 orthogonal cursors (Temporal, Source, Gesture, Topic, People), fires parallel Pieces LTM searches with pagination until confidence >= threshold, then orchestrates the full neuro (10 iter) -> code-hardener (5 iter) -> implementation -> lint-fixer (5 loop) chain. Self-evolutionary: after every run, analyzes outcomes and updates its own structure. Integrates, wraps, and supersedes neuro, code-hardener, and lint-fixer.
Mandatory external architectural and code-review harness for ANY non-trivial task including code changes, bug fixes, new features, refactors, API changes, data contract changes, performance optimizations, security fixes, architectural decisions, configuration changes, test modifications, integration work, debugging, analysis, planning, and implementation. Use when the user requests any modification, analysis, debugging, testing, integration, or planning work. Use for EVERY prompt that involves understanding or changing code, data, or configuration. When in doubt, ALWAYS use this skill. Skip only for trivial changes: typo fixes, formatting-only changes, one-line lint fixes, comment cleanup, simple unused import removal.
Mandatory second-stage logic filter and architectural hardening skill that runs after NEURO for ANY code change, bug fix, new feature, refactor, API change, data contract change, performance optimization, security fix, architectural decision, configuration change, test modification, integration work, debugging, analysis, planning, or implementation. Executes exactly 5 mandatory iterations. Validates NEURO recommendations against repo truth, calls NEURO with filtered critique, then emits the only implementation plan opencode may follow. Use after any neuro skill usage. Use when code is about to be edited. Use for every non-trivial code modification.
Mandatory post-implementation lint and type-checking skill that runs after ANY code change, bug fix, new feature, refactor, API change, or configuration modification. Ensures all ruff lint errors, mypy type errors, and ESLint issues are resolved. Use after every implementation that modifies source files. Use when code has been edited. Use for every response where file edits were made.
Mandatory second-stage logic filter and architectural hardening skill that runs after NEURO. It validates NEURO recommendations against repo truth, calls NEURO a second time with filtered critique, then emits the only implementation plan Gemini may follow.
| name | neuro |
| description | Mandatory external architectural and code-review harness for non-trivial code changes, PR reviews, refactors, modeling logic, dashboard/API integration, performance-sensitive work, and implementation planning. Must be followed by the code-hardener skill before Gemini edits production code. |
neuro routes non-trivial engineering work through NEURO / Neurometric before Gemini makes code changes.
Gemini is not the first architectural authority. NEURO provides first-pass architectural review, risk analysis, minimal patch strategy, edge cases, and validation guidance.
NEURO is still advisory. It is not final truth. The current repository, tests, PROJECT_CONTEXT.md, and user instructions remain authoritative.
For non-trivial work, this skill is always step 1 of a two-skill chain:
neuro → code-hardener → Gemini implementation.
Use this skill for:
Skip only for trivial changes:
If unsure, use NEURO.
Before calling NEURO, gather:
PROJECT_CONTEXT.mdRequest NEURO to return:
After NEURO returns output, Gemini must not edit source code.
Gemini must activate code-hardener and pass the NEURO output into it.
The next required artifact is:
.neuro/chains/latest/04_final_code_hardener_plan.md
No production source file may be edited until that artifact exists and matches the current user request.
If live NEURO is unavailable:
Keep this SKILL.md concise. Detailed workflow is in:
workflows/neuro-review-protocol.mdExecutable harness is in:
scripts/neuro_harness.py